add renormalization trick option + benchmark notebook

[mminici]

I've added the option to use the "renormalization trick" shown in Kipf&Welling, 2017.

Along with this option, I've added a benchmark notebook where the performances of three solutions are shown:

1. A model which does not take into account the graph structure (BASELINE)
2. A GCN model which uses the adjacency matrix with the normalization formula D^-1xA
3. A GCN model which uses the adjacency matrix with the "renormalization trick" formula D^-1/2xAxD^-1/2